X-Git-Url: https://bearssl.org/gitweb//home/git/?p=BearSSL;a=blobdiff_plain;f=src%2Finner.h;h=6095806ebabb700a457dd8b925d98622550ca163;hp=d2cc4c9f6ee02c7b5e9d99e9aa9a2c13fab552a9;hb=90bc9406c31e03d09b3d835c3cbabfec83f4e94d;hpb=b42bd5972f935ffc32019acac6f8a07ae08ae9c2;ds=sidebyside diff --git a/src/inner.h b/src/inner.h index d2cc4c9..6095806 100644 --- a/src/inner.h +++ b/src/inner.h @@ -31,6 +31,15 @@ #include "config.h" #include "bearssl.h" +/* + * On MSVC, disable the warning about applying unary minus on an + * unsigned type: it is standard, we do it all the time, and for + * good reasons. + */ +#if _MSC_VER +#pragma warning( disable : 4146 ) +#endif + /* * Maximum size for a RSA modulus (in bits). Allocated stack buffers * depend on that size, so this value should be kept small. Currently, @@ -96,58 +105,238 @@ #endif #endif +/* + * Set BR_LOMUL on platforms where it makes sense. + */ +#ifndef BR_LOMUL +#if BR_ARMEL_CORTEX_GCC +#define BR_LOMUL 1 +#endif +#endif + +/* + * Determine whether x86 AES instructions are understood by the compiler. + */ +#ifndef BR_AES_X86NI +#if (__i386__ || __x86_64__) \ + && ((__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) \ + || (__clang_major__ > 3 \ + || (__clang_major__ == 3 && __clang_minor__ >= 7))) +#define BR_AES_X86NI 1 +#elif (_M_IX86 || _M_X64) && (_MSC_VER >= 1700) +#define BR_AES_X86NI 1 +#endif +#endif + +/* + * If we use x86 AES instruction, determine the compiler brand. + */ +#if BR_AES_X86NI +#ifndef BR_AES_X86NI_GCC +#if __GNUC__ +#define BR_AES_X86NI_GCC 1 +#endif +#endif +#ifndef BR_AES_X86NI_MSC +#if _MSC_VER >= 1700 +#define BR_AES_X86NI_MSC 1 +#endif +#endif +#endif + +/* + * A macro to tag a function with a "target" attribute (for GCC and Clang). + */ +#if BR_AES_X86NI_GCC +#define BR_TARGET(x) __attribute__((target(x))) +#else +#define BR_TARGET(x) +#endif + +/* + * GCC versions from 4.4 to 4.8 (inclusive) must use a special #pragma + * to activate extra opcodes before including the relevant intrinsic + * headers. But these don't work with Clang (which does not need them + * either). + */ +#if BR_AES_X86NI_GCC && !defined BR_AES_X86NI_GCC_OLD +#if __GNUC__ == 4 && __GNUC_MINOR__ >= 4 && __GNUC_MINOR__ <= 8 && !__clang__ +#define BR_AES_X86NI_GCC_OLD 1 +#endif +#endif + +/* + * POWER8 crypto support. We rely on compiler macros for the + * architecture, since we do not have a reliable, simple way to detect + * the required support at runtime (we could try running an opcode, and + * trapping the exception or signal on illegal instruction, but this + * induces some non-trivial OS dependencies that we would prefer to + * avoid if possible). + */ +#ifndef BR_POWER8 +#if __GNUC__ && ((_ARCH_PWR8 || _ARCH_PPC) && __CRYPTO__) +#define BR_POWER8 1 +#endif +#endif + +/* + * Detect endinanness on POWER8. + */ +#if BR_POWER8 +#if defined BR_POWER8_LE +#undef BR_POWER8_BE +#if BR_POWER8_LE +#define BR_POWER8_BE 0 +#else +#define BR_POWER8_BE 1 +#endif +#elif defined BR_POWER8_BE +#undef BR_POWER8_LE +#if BR_POWER8_BE +#define BR_POWER8_LE 0 +#else +#define BR_POWER8_LE 1 +#endif +#else +#if __LITTLE_ENDIAN__ +#define BR_POWER8_LE 1 +#define BR_POWER8_BE 0 +#else +#define BR_POWER8_LE 0 +#define BR_POWER8_BE 1 +#endif +#endif +#endif + +/* + * Detect support for 128-bit integers. + */ +#if !defined BR_INT128 && !defined BR_UMUL128 +#ifdef __SIZEOF_INT128__ +#define BR_INT128 1 +#elif _M_X64 +#define BR_UMUL128 1 +#endif +#endif + +/* + * Detect support for unaligned accesses with known endianness. + * + * x86 (both 32-bit and 64-bit) is little-endian and allows unaligned + * accesses. + * + * POWER/PowerPC allows unaligned accesses when big-endian. POWER8 and + * later also allow unaligned accesses when little-endian. + */ +#if !defined BR_LE_UNALIGNED && !defined BR_BE_UNALIGNED + +#if __i386 || __i386__ || __x86_64__ || _M_IX86 || _M_X64 +#define BR_LE_UNALIGNED 1 +#elif BR_POWER8_BE +#define BR_BE_UNALIGNED 1 +#elif BR_POWER8_LE +#define BR_LE_UNALIGNED 1 +#elif (__powerpc__ || __powerpc64__ || _M_PPC || _ARCH_PPC || _ARCH_PPC64) \\ + && __BIG_ENDIAN__ +#define BR_BE_UNALIGNED 1 +#endif + +#endif + /* ==================================================================== */ /* * Encoding/decoding functions. * * 32-bit and 64-bit decoding, both little-endian and big-endian, is - * implemented with the inline functions below. These functions are - * generic: they don't depend on the architecture natural endianness, - * and they can handle unaligned accesses. Optimized versions for some - * specific architectures may be implemented at a later time. - */ + * implemented with the inline functions below. + * + * When allowed by some compile-time options (autodetected or provided), + * optimised code is used, to perform direct memory access when the + * underlying architecture supports it, both for endianness and + * alignment. This, however, may trigger strict aliasing issues; the + * code below uses unions to perform (supposedly) safe type punning. + * Since the C aliasing rules are relatively complex and were amended, + * or at least re-explained with different phrasing, in all successive + * versions of the C standard, it is always a bit risky to bet that any + * specific version of a C compiler got it right, for some notion of + * "right". + */ + +typedef union { + uint16_t u; + unsigned char b[sizeof(uint16_t)]; +} br_union_u16; + +typedef union { + uint32_t u; + unsigned char b[sizeof(uint32_t)]; +} br_union_u32; + +typedef union { + uint64_t u; + unsigned char b[sizeof(uint64_t)]; +} br_union_u64; static inline void br_enc16le(void *dst, unsigned x) { +#if BR_LE_UNALIGNED + ((br_union_u16 *)dst)->u = x; +#else unsigned char *buf; buf = dst; buf[0] = (unsigned char)x; buf[1] = (unsigned char)(x >> 8); +#endif } static inline void br_enc16be(void *dst, unsigned x) { +#if BR_BE_UNALIGNED + ((br_union_u16 *)dst)->u = x; +#else unsigned char *buf; buf = dst; buf[0] = (unsigned char)(x >> 8); buf[1] = (unsigned char)x; +#endif } static inline unsigned br_dec16le(const void *src) { +#if BR_LE_UNALIGNED + return ((const br_union_u16 *)src)->u; +#else const unsigned char *buf; buf = src; return (unsigned)buf[0] | ((unsigned)buf[1] << 8); +#endif } static inline unsigned br_dec16be(const void *src) { +#if BR_BE_UNALIGNED + return ((const br_union_u16 *)src)->u; +#else const unsigned char *buf; buf = src; return ((unsigned)buf[0] << 8) | (unsigned)buf[1]; +#endif } static inline void br_enc32le(void *dst, uint32_t x) { +#if BR_LE_UNALIGNED + ((br_union_u32 *)dst)->u = x; +#else unsigned char *buf; buf = dst; @@ -155,11 +344,15 @@ br_enc32le(void *dst, uint32_t x) buf[1] = (unsigned char)(x >> 8); buf[2] = (unsigned char)(x >> 16); buf[3] = (unsigned char)(x >> 24); +#endif } static inline void br_enc32be(void *dst, uint32_t x) { +#if BR_BE_UNALIGNED + ((br_union_u32 *)dst)->u = x; +#else unsigned char *buf; buf = dst; @@ -167,11 +360,15 @@ br_enc32be(void *dst, uint32_t x) buf[1] = (unsigned char)(x >> 16); buf[2] = (unsigned char)(x >> 8); buf[3] = (unsigned char)x; +#endif } static inline uint32_t br_dec32le(const void *src) { +#if BR_LE_UNALIGNED + return ((const br_union_u32 *)src)->u; +#else const unsigned char *buf; buf = src; @@ -179,11 +376,15 @@ br_dec32le(const void *src) | ((uint32_t)buf[1] << 8) | ((uint32_t)buf[2] << 16) | ((uint32_t)buf[3] << 24); +#endif } static inline uint32_t br_dec32be(const void *src) { +#if BR_BE_UNALIGNED + return ((const br_union_u32 *)src)->u; +#else const unsigned char *buf; buf = src; @@ -191,46 +392,63 @@ br_dec32be(const void *src) | ((uint32_t)buf[1] << 16) | ((uint32_t)buf[2] << 8) | (uint32_t)buf[3]; +#endif } static inline void br_enc64le(void *dst, uint64_t x) { +#if BR_LE_UNALIGNED + ((br_union_u64 *)dst)->u = x; +#else unsigned char *buf; buf = dst; br_enc32le(buf, (uint32_t)x); br_enc32le(buf + 4, (uint32_t)(x >> 32)); +#endif } static inline void br_enc64be(void *dst, uint64_t x) { +#if BR_BE_UNALIGNED + ((br_union_u64 *)dst)->u = x; +#else unsigned char *buf; buf = dst; br_enc32be(buf, (uint32_t)(x >> 32)); br_enc32be(buf + 4, (uint32_t)x); +#endif } static inline uint64_t br_dec64le(const void *src) { +#if BR_LE_UNALIGNED + return ((const br_union_u64 *)src)->u; +#else const unsigned char *buf; buf = src; return (uint64_t)br_dec32le(buf) | ((uint64_t)br_dec32le(buf + 4) << 32); +#endif } static inline uint64_t br_dec64be(const void *src) { +#if BR_BE_UNALIGNED + return ((const br_union_u64 *)src)->u; +#else const unsigned char *buf; buf = src; return ((uint64_t)br_dec32be(buf) << 32) | (uint64_t)br_dec32be(buf + 4); +#endif } /* @@ -300,7 +518,7 @@ static inline void br_multihash_copyimpl(br_multihash_context *dst, const br_multihash_context *src) { - memcpy(dst->impl, src->impl, sizeof src->impl); + memcpy((void *)dst->impl, src->impl, sizeof src->impl); } /* ==================================================================== */ @@ -532,11 +750,30 @@ MAX(uint32_t x, uint32_t y) * (old) platforms where the default MUL31 is not. Unfortunately, it is * also substantially slower, and yields larger code, on more modern * platforms, which is why it is deactivated by default. + * + * MUL31_lo() must do some extra work because on some platforms, the + * _signed_ multiplication may return early if the top bits are 1. + * Simply truncating (casting) the output of MUL31() would not be + * sufficient, because the compiler may notice that we keep only the low + * word, and then replace automatically the unsigned multiplication with + * a signed multiplication opcode. */ #define MUL31(x, y) ((uint64_t)((x) | (uint32_t)0x80000000) \ * (uint64_t)((y) | (uint32_t)0x80000000) \ - ((uint64_t)(x) << 31) - ((uint64_t)(y) << 31) \ - ((uint64_t)1 << 62)) +static inline uint32_t +MUL31_lo(uint32_t x, uint32_t y) +{ + uint32_t xl, xh; + uint32_t yl, yh; + + xl = (x & 0xFFFF) | (uint32_t)0x80000000; + xh = (x >> 16) | (uint32_t)0x80000000; + yl = (y & 0xFFFF) | (uint32_t)0x80000000; + yh = (y >> 16) | (uint32_t)0x80000000; + return (xl * yl + ((xl * yh + xh * yl) << 16)) & (uint32_t)0x7FFFFFFF; +} #else @@ -544,9 +781,49 @@ MAX(uint32_t x, uint32_t y) * Multiply two 31-bit integers, with a 62-bit result. This default * implementation assumes that the basic multiplication operator * yields constant-time code. + * The MUL31_lo() macro returns only the low 31 bits of the product. */ -#define MUL31(x, y) ((uint64_t)(x) * (uint64_t)(y)) +#define MUL31(x, y) ((uint64_t)(x) * (uint64_t)(y)) +#define MUL31_lo(x, y) (((uint32_t)(x) * (uint32_t)(y)) & (uint32_t)0x7FFFFFFF) + +#endif + +/* + * Multiply two words together; the sum of the lengths of the two + * operands must not exceed 31 (for instance, one operand may use 16 + * bits if the other fits on 15). If BR_CT_MUL15 is non-zero, then the + * macro will contain some extra operations that help in making the + * operation constant-time on some platforms, where the basic 32-bit + * multiplication is not constant-time. + */ +#if BR_CT_MUL15 +#define MUL15(x, y) (((uint32_t)(x) | (uint32_t)0x80000000) \ + * ((uint32_t)(y) | (uint32_t)0x80000000) \ + & (uint32_t)0x7FFFFFFF) +#else +#define MUL15(x, y) ((uint32_t)(x) * (uint32_t)(y)) +#endif +/* + * Arithmetic right shift (sign bit is copied). What happens when + * right-shifting a negative value is _implementation-defined_, so it + * does not trigger undefined behaviour, but it is still up to each + * compiler to define (and document) what it does. Most/all compilers + * will do an arithmetic shift, the sign bit being used to fill the + * holes; this is a native operation on the underlying CPU, and it would + * make little sense for the compiler to do otherwise. GCC explicitly + * documents that it follows that convention. + * + * Still, if BR_NO_ARITH_SHIFT is defined (and non-zero), then an + * alternate version will be used, that does not rely on such + * implementation-defined behaviour. Unfortunately, it is also slower + * and yields bigger code, which is why it is deactivated by default. + */ +#if BR_NO_ARITH_SHIFT +#define ARSH(x, n) (((uint32_t)(x) >> (n)) \ + | ((-((uint32_t)(x) >> 31)) << (32 - (n)))) +#else +#define ARSH(x, n) ((*(int32_t *)&(x)) >> (n)) #endif /* @@ -1009,6 +1286,60 @@ void br_i31_mulacc(uint32_t *d, const uint32_t *a, const uint32_t *b); /* ==================================================================== */ +/* + * FIXME: document "i15" functions. + */ + +static inline void +br_i15_zero(uint16_t *x, uint16_t bit_len) +{ + *x ++ = bit_len; + memset(x, 0, ((bit_len + 15) >> 4) * sizeof *x); +} + +uint32_t br_i15_iszero(const uint16_t *x); + +uint16_t br_i15_ninv15(uint16_t x); + +uint32_t br_i15_add(uint16_t *a, const uint16_t *b, uint32_t ctl); + +uint32_t br_i15_sub(uint16_t *a, const uint16_t *b, uint32_t ctl); + +void br_i15_muladd_small(uint16_t *x, uint16_t z, const uint16_t *m); + +void br_i15_montymul(uint16_t *d, const uint16_t *x, const uint16_t *y, + const uint16_t *m, uint16_t m0i); + +void br_i15_to_monty(uint16_t *x, const uint16_t *m); + +void br_i15_modpow(uint16_t *x, const unsigned char *e, size_t elen, + const uint16_t *m, uint16_t m0i, uint16_t *t1, uint16_t *t2); + +uint32_t br_i15_modpow_opt(uint16_t *x, const unsigned char *e, size_t elen, + const uint16_t *m, uint16_t m0i, uint16_t *tmp, size_t twlen); + +void br_i15_encode(void *dst, size_t len, const uint16_t *x); + +uint32_t br_i15_decode_mod(uint16_t *x, + const void *src, size_t len, const uint16_t *m); + +void br_i15_rshift(uint16_t *x, int count); + +uint32_t br_i15_bit_length(uint16_t *x, size_t xlen); + +void br_i15_decode(uint16_t *x, const void *src, size_t len); + +void br_i15_from_monty(uint16_t *x, const uint16_t *m, uint16_t m0i); + +void br_i15_decode_reduce(uint16_t *x, + const void *src, size_t len, const uint16_t *m); + +void br_i15_reduce(uint16_t *x, const uint16_t *a, const uint16_t *m); + +void br_i15_mulacc(uint16_t *d, const uint16_t *a, const uint16_t *b); + +/* ==================================================================== */ + static inline size_t br_digest_size(const br_hash_class *digest_class) { @@ -1306,6 +1637,63 @@ unsigned br_aes_ct64_keysched(uint64_t *comp_skey, void br_aes_ct64_skey_expand(uint64_t *skey, unsigned num_rounds, const uint64_t *comp_skey); +/* + * Test support for AES-NI opcodes. + */ +int br_aes_x86ni_supported(void); + +/* + * AES key schedule, using x86 AES-NI instructions. This yields the + * subkeys in the encryption direction. Number of rounds is returned. + * Key size MUST be 16, 24 or 32 bytes; otherwise, 0 is returned. + */ +unsigned br_aes_x86ni_keysched_enc(unsigned char *skni, + const void *key, size_t len); + +/* + * AES key schedule, using x86 AES-NI instructions. This yields the + * subkeys in the decryption direction. Number of rounds is returned. + * Key size MUST be 16, 24 or 32 bytes; otherwise, 0 is returned. + */ +unsigned br_aes_x86ni_keysched_dec(unsigned char *skni, + const void *key, size_t len); + +/* + * Test support for AES POWER8 opcodes. + */ +int br_aes_pwr8_supported(void); + +/* + * AES key schedule, using POWER8 instructions. This yields the + * subkeys in the encryption direction. Number of rounds is returned. + * Key size MUST be 16, 24 or 32 bytes; otherwise, 0 is returned. + */ +unsigned br_aes_pwr8_keysched(unsigned char *skni, + const void *key, size_t len); + +/* ==================================================================== */ +/* + * RSA. + */ + +/* + * Apply proper PKCS#1 v1.5 padding (for signatures). 'hash_oid' is + * the encoded hash function OID, or NULL. + */ +uint32_t br_rsa_pkcs1_sig_pad(const unsigned char *hash_oid, + const unsigned char *hash, size_t hash_len, + uint32_t n_bitlen, unsigned char *x); + +/* + * Check PKCS#1 v1.5 padding (for signatures). 'hash_oid' is the encoded + * hash function OID, or NULL. The provided 'sig' value is _after_ the + * modular exponentiation, i.e. it should be the padded hash. On + * success, the hashed message is extracted. + */ +uint32_t br_rsa_pkcs1_sig_unpad(const unsigned char *sig, size_t sig_len, + const unsigned char *hash_oid, size_t hash_len, + unsigned char *hash_out); + /* ==================================================================== */ /* * Elliptic curves. @@ -1328,24 +1716,13 @@ extern const br_ec_curve_def br_secp384r1; extern const br_ec_curve_def br_secp521r1; /* - * Type for the parameters for a "prime curve": - * coordinates are in GF(p), with p prime - * curve equation is Y^2 = X^3 - 3*X + b - * b is in Montgomery representation - * curve order is n and is prime - * base point is G (encoded) and has order n + * For Curve25519, the advertised "order" really is 2^255-1, since the + * point multipliction function really works over arbitrary 255-bit + * scalars. This value is only meant as a hint for ECDH key generation; + * only ECDSA uses the exact curve order, and ECDSA is not used with + * that specific curve. */ -typedef struct { - const uint32_t *p; - const uint32_t *b; - const uint32_t p0i; -} br_ec_prime_i31_curve; - -extern const br_ec_prime_i31_curve br_ec_prime_i31_secp256r1; -extern const br_ec_prime_i31_curve br_ec_prime_i31_secp384r1; -extern const br_ec_prime_i31_curve br_ec_prime_i31_secp521r1; - -#define BR_EC_I31_LEN ((BR_MAX_EC_SIZE + 61) / 31) +extern const br_ec_curve_def br_curve25519; /* * Decode some bytes as an i31 integer, with truncation (corresponding @@ -1357,6 +1734,16 @@ extern const br_ec_prime_i31_curve br_ec_prime_i31_secp521r1; void br_ecdsa_i31_bits2int(uint32_t *x, const void *src, size_t len, uint32_t ebitlen); +/* + * Decode some bytes as an i15 integer, with truncation (corresponding + * to the 'bits2int' operation in RFC 6979). The target ENCODED bit + * length is provided as last parameter. The resulting value will have + * this declared bit length, and consists the big-endian unsigned decoding + * of exactly that many bits in the source (capped at the source length). + */ +void br_ecdsa_i15_bits2int(uint16_t *x, + const void *src, size_t len, uint32_t ebitlen); + /* ==================================================================== */ /* * SSL/TLS support functions. @@ -1561,4 +1948,85 @@ int br_ssl_choose_hash(unsigned bf); /* ==================================================================== */ +/* + * PowerPC / POWER assembly stuff. The special BR_POWER_ASM_MACROS macro + * must be defined before including this file; this is done by source + * files that use some inline assembly for PowerPC / POWER machines. + */ + +#if BR_POWER_ASM_MACROS + +#define lxvw4x(xt, ra, rb) lxvw4x_(xt, ra, rb) +#define stxvw4x(xt, ra, rb) stxvw4x_(xt, ra, rb) + +#define bdnz(foo) bdnz_(foo) +#define beq(foo) beq_(foo) + +#define li(rx, value) li_(rx, value) +#define addi(rx, ra, imm) addi_(rx, ra, imm) +#define cmpldi(rx, imm) cmpldi_(rx, imm) +#define mtctr(rx) mtctr_(rx) +#define vspltb(vrt, vrb, uim) vspltb_(vrt, vrb, uim) +#define vspltw(vrt, vrb, uim) vspltw_(vrt, vrb, uim) +#define vspltisb(vrt, imm) vspltisb_(vrt, imm) +#define vspltisw(vrt, imm) vspltisw_(vrt, imm) +#define vrlw(vrt, vra, vrb) vrlw_(vrt, vra, vrb) +#define vsbox(vrt, vra) vsbox_(vrt, vra) +#define vxor(vrt, vra, vrb) vxor_(vrt, vra, vrb) +#define vand(vrt, vra, vrb) vand_(vrt, vra, vrb) +#define vsro(vrt, vra, vrb) vsro_(vrt, vra, vrb) +#define vsl(vrt, vra, vrb) vsl_(vrt, vra, vrb) +#define vsldoi(vt, va, vb, sh) vsldoi_(vt, va, vb, sh) +#define vsr(vrt, vra, vrb) vsr_(vrt, vra, vrb) +#define vadduwm(vrt, vra, vrb) vadduwm_(vrt, vra, vrb) +#define vsububm(vrt, vra, vrb) vsububm_(vrt, vra, vrb) +#define vsubuwm(vrt, vra, vrb) vsubuwm_(vrt, vra, vrb) +#define vsrw(vrt, vra, vrb) vsrw_(vrt, vra, vrb) +#define vcipher(vt, va, vb) vcipher_(vt, va, vb) +#define vcipherlast(vt, va, vb) vcipherlast_(vt, va, vb) +#define vncipher(vt, va, vb) vncipher_(vt, va, vb) +#define vncipherlast(vt, va, vb) vncipherlast_(vt, va, vb) +#define vperm(vt, va, vb, vc) vperm_(vt, va, vb, vc) +#define vpmsumd(vt, va, vb) vpmsumd_(vt, va, vb) +#define xxpermdi(vt, va, vb, d) xxpermdi_(vt, va, vb, d) + +#define lxvw4x_(xt, ra, rb) "\tlxvw4x\t" #xt "," #ra "," #rb "\n" +#define stxvw4x_(xt, ra, rb) "\tstxvw4x\t" #xt "," #ra "," #rb "\n" + +#define label(foo) #foo "%=:\n" +#define bdnz_(foo) "\tbdnz\t" #foo "%=\n" +#define beq_(foo) "\tbeq\t" #foo "%=\n" + +#define li_(rx, value) "\tli\t" #rx "," #value "\n" +#define addi_(rx, ra, imm) "\taddi\t" #rx "," #ra "," #imm "\n" +#define cmpldi_(rx, imm) "\tcmpldi\t" #rx "," #imm "\n" +#define mtctr_(rx) "\tmtctr\t" #rx "\n" +#define vspltb_(vrt, vrb, uim) "\tvspltb\t" #vrt "," #vrb "," #uim "\n" +#define vspltw_(vrt, vrb, uim) "\tvspltw\t" #vrt "," #vrb "," #uim "\n" +#define vspltisb_(vrt, imm) "\tvspltisb\t" #vrt "," #imm "\n" +#define vspltisw_(vrt, imm) "\tvspltisw\t" #vrt "," #imm "\n" +#define vrlw_(vrt, vra, vrb) "\tvrlw\t" #vrt "," #vra "," #vrb "\n" +#define vsbox_(vrt, vra) "\tvsbox\t" #vrt "," #vra "\n" +#define vxor_(vrt, vra, vrb) "\tvxor\t" #vrt "," #vra "," #vrb "\n" +#define vand_(vrt, vra, vrb) "\tvand\t" #vrt "," #vra "," #vrb "\n" +#define vsro_(vrt, vra, vrb) "\tvsro\t" #vrt "," #vra "," #vrb "\n" +#define vsl_(vrt, vra, vrb) "\tvsl\t" #vrt "," #vra "," #vrb "\n" +#define vsldoi_(vt, va, vb, sh) "\tvsldoi\t" #vt "," #va "," #vb "," #sh "\n" +#define vsr_(vrt, vra, vrb) "\tvsr\t" #vrt "," #vra "," #vrb "\n" +#define vadduwm_(vrt, vra, vrb) "\tvadduwm\t" #vrt "," #vra "," #vrb "\n" +#define vsububm_(vrt, vra, vrb) "\tvsububm\t" #vrt "," #vra "," #vrb "\n" +#define vsubuwm_(vrt, vra, vrb) "\tvsubuwm\t" #vrt "," #vra "," #vrb "\n" +#define vsrw_(vrt, vra, vrb) "\tvsrw\t" #vrt "," #vra "," #vrb "\n" +#define vcipher_(vt, va, vb) "\tvcipher\t" #vt "," #va "," #vb "\n" +#define vcipherlast_(vt, va, vb) "\tvcipherlast\t" #vt "," #va "," #vb "\n" +#define vncipher_(vt, va, vb) "\tvncipher\t" #vt "," #va "," #vb "\n" +#define vncipherlast_(vt, va, vb) "\tvncipherlast\t" #vt "," #va "," #vb "\n" +#define vperm_(vt, va, vb, vc) "\tvperm\t" #vt "," #va "," #vb "," #vc "\n" +#define vpmsumd_(vt, va, vb) "\tvpmsumd\t" #vt "," #va "," #vb "\n" +#define xxpermdi_(vt, va, vb, d) "\txxpermdi\t" #vt "," #va "," #vb "," #d "\n" + +#endif + +/* ==================================================================== */ + #endif